When the terrain height batch loses a chunk to the rate limit, the cameras
in it keep a sea-level fallback height. In the mountains that put them far
under the surface, so the view pass judged them by a point that was not
where they are drawn: a tilted view dropped the count marker, a click on
the marker framed empty ground, and icons drifted on tilt and pan.
- A camera with no real ground height takes the height of the terrain the
globe draws once it is near the screen (up to 300 reads a pass, through
the height service), and a later answer from the batch replaces it.
- A count marker is anchored to the drawn terrain under its centre, and a
click frames that ground.
- Cameras on one mount draw side by side, 30 px apart, each with its own
callout, instead of as one icon with several callouts.
The photoreal map stack is not covered: terrain is not the surface there.
Cards, the geometry queue, the nearest-camera search, auto-hop's view
context and the view motion pass read where the eye is from the camera
port's pose, through one model helper that answers null when there is no
camera or the eye has no geographic position, as the code did for a
missing cartographic position. The motion pass measures with geo/vector,
and a camera's first position comes from geo/ellipsoid as a plain point;
everything that reads it takes x, y and z. Navigation flies through the
port it already used, now found through rendererOf; the rendering part
loads the camera factory so a viewer of a caller's own still has one.
Cards, geometryQueue, navigation and lifecycle leave the allow-list.
The CCTV layer kept every loaded camera live: each view change touched every
icon, every notification rebuilt every camera's state, and the picker listed
the whole catalog. With the default 4,000 cameras, a continental drag ran at
152-179 fps with 18-24 frames over 30 ms (worst 112 ms). It now runs at
235 fps with none. The layer now:
- shows icons only on or near the screen, and when zoomed out keeps one camera
per world-anchored cell (healthy feeds first), so panning keeps the same
cameras;
- grounds and draws coverage only for the cameras in view, the active camera
and its coverage neighbours, retrying a camera at most twice per view regime;
- reruns the view pass while the camera moves, paced by its own cost, and
projects with one matrix per pass instead of a Cesium call per camera;
- keeps the billboard buffer-usage hint fixed, so showing or hiding a few
icons no longer rewrites the vertex data of every billboard;
- lists the cameras in view (active camera first) in the picker, closing with
the count of cameras outside the view, and builds the full camera list in
the UI state only when it is read.